Ensuring the sustainability of PAs requires information on the effectiveness of management decisions at large spatial scales (Leverington et al. 2010; Rodrigues & Cazalis, 2020). However, few PAs have established systems to evaluate management effectiveness or to determine whether they are achieving their aims such as conserving biodiversity (Hockings et al. 2000; Parirish et al. 2003). For example, in South African state PAs the Management Effectiveness Tracking Tool (METT) is used to evaluate management effectiveness. The tool monitors if management actions take place, but do not report on the effect(s) of these i.e. the METT would check if a planned block burn took place, but does not measure the biodiversity outcomes of the block burn. Moreover, most monitoring programs are opportunistic rather than strategic, mostly because it is unclear what should be monitored, where, and why (Dale and Beyeler 2001). This is particularly important given the effect of the COVID-19 pandemic on PA funding and planning for biodiversity conservation in a post-COVID-19 economy (Sandbrook et al. 2020). Here, we propose a cost-effective and standardised way for PA managers to monitor changes in heterogeneity directly and across scales. By monitoring heterogeneity, park managers can answer logical questions concerning environmental change that could affect species diversity and ecosystem function. For example, long-term trends in heterogeneity might reveal differences among PAs that can be linked to management practices, such as the opening and closing of waterholes or changes in fire regimes. Moreover, drawing from the maps for each PA, managers may be able to identify possible “hotspots” and focus research into the empirical effects of changes in heterogeneity, for example associated changes in species diversity and potential ecosystem regime shifts. Such maps may also prove useful to detect changes in vegetation structure and composition, which could lead to an alteration in the provision of ecosystem services or indicate local extinction events.
Monitoring heterogeneity within numerous PAs may appear to be a daunting task. However, it does not involve conservation planners, expert knowledge, or the investment of additional field resources – the heterogeneity product for each PA can be regularly updated by replacing the acquisition dates in the associated R scripts and re-running the remaining code-parts (Supporting Information S2 and S3). The approach is generic and applicable to different datasets at all scales (local, regional, national) and at different resolutions. This framework now exists for the 41 areas included in this analyses but can be extended to include every PA nationally, or even globally, and the data product fully automated. For this study, we used the Mann-Kendall test to determine significant trends in heterogeneity, however other models can also be fitted to the time series. For example, piece-wise regression can be used to detect breakpoints in the time series, which could signify the impact of management interventions. Importantly, we do not advocate that this monitoring approach replace current monitoring programs – rather that it can add value and further support the monitoring of PA effectiveness in achieving desirable conservation outcomes.
Are PAs in South Africa successful in maintaining heterogeneity? Generally, yes. For the majority (63%) of PAs included in this analysis, heterogeneity did not change significantly during the last 28 years. This does not mean that heterogeneity remained unchanged across the entire PA, but rather that the mean heterogeneity for these PAs remained unchanged (within each PA there were areas where heterogeneity increased, decreased or remained stable). PAs where heterogeneity remained stable were found across the country, but noticeably, almost all of the PAs in the subtropical low veld Kruger region were stable. Many of these PAs are part of an open system with the Kruger National Park, South Africa’s largest conservation area. Here fences were removed in the early 1990’s and since then the area has functioned as a relatively open system, although management practices do differ among the different entities. The management philosophy of South African National Parks (SANParks) is centred on maintaining heterogeneity, where the restoration of ecological processes should result in the formation of more heterogeneous landscapes. Our results suggest that in the KNP, trends in heterogeneity since 1990 were stable across the three different regions of the park: north, central, and south, which indicate that management actions are having the desired effect of maintaining heterogeneity. Moreover, in the south and central regions, heterogeneity increased for the majority of pixels, which is a further positive development from a management perspective. However, in the north, the majority of pixels decreased, although the general trend was stable. This may indicate a possible shift towards a more homogenous landscape, possibly in response to changes in elephant space use (MacFadyen et al. 2019), rainfall (Smit et al. 2013), or frequent fires (Ribeiro et al. 2019). The maps presented here could aid managers in identifying where these vulnerable areas are, as well as determine and mitigate the potential drivers of decreasing heterogeneity.
There was pronounced spatial variation in heterogeneity within PAs. Specifically, the proportion of pixels that increased, decreased, or remained stable were never more than half of the total number of pixels. This suggests that the overall landscape complexity associated with PAs, where underlying physical landscape templates (e.g., topography, soils, watershed areas) interact with dynamic landscape processes and stochastic disturbance events generate a ‘heterogeneity mosaic’, irrespective of management actions or regional factors (e.g. rainfall) (see MacFadyen et al. 2016). This can be exploited by managers by stratifying PA monitoring and research planning using these maps as blueprints. For example, it might be easier and more cost effective to maintain heterogeneity in more complex landscapes than artificially increasing heterogeneity in more simple areas. This is important, particularly given that heterogeneity is noticeably higher along roads, fences, infrastructure, and waterholes, and shows that a significant increase in heterogeneity over time is not always caused by natural phenomena. Local factors (e.g., roads, waterholes, or herbivory) can therefore be manipulated to some extent to alter heterogeneity at fine scales. However, deciding where to do this is crucial for PA management as such an increase in heterogeneity might have the opposite effect of boosting ecological integrity by increasing edge effects, eliminating the seasonal ranges of animals, and destroying habitats.
Although heterogeneity remained stable for most of the PAs considered in our analysis, there were also 11 (32%) PAs where heterogeneity decreased significantly during the last three decades. Closer inspection revealed that heterogeneity also decreased for the majority of pixels in 10 of these 11 PAs. These trends are worrying as a decrease in heterogeneity may indicate a decrease in species richness and the viability of animal populations that occur here (e.g., McCleery et al. 2018). Unravelling the drivers of changes in heterogeneity are beyond the scope of this contribution, yet we know that in African savannahs fire, rainfall, and herbivores play key roles in shaping vegetation communities (Sinclair & Walker 2003). To maintain heterogeneity, animals must utilise distinct wet and dry season ranges, fires should occur sporadically and unpredictably, and woody encroachment must be inhibited by local variations in rainfall, soil type, grazing pressure, and even the impact of people (Wang et al. 2013; Smit et al. 2013; Van Langevelde et al. 2017). However, in fenced reserves with ecologically arbitrary boundaries and set management regimes these processes may be fundamentally altered, which could reduce heterogeneity. For example, in some PAs, the widespread provision of artificial waterholes may degrade functional wet and dry season resources by homogenising animal distribution patterns. Moreover, uniform fire regimes can result in more homogenous vegetation structures where small trees and shrubs expand at the expense of taller savannah woodlands (van Wilgen et al. 2003), while CO2 levels may be driving woody encroachment (Stevens et al. 2016). We may therefore be recording a type of ‘park effect’ where herbivory, fire, and a combination of rainfall and increasing CO2 levels may be contributing to the homogenization of these PAs (see also Guldemond & van Aarde, 2010). This should be of particular concern to PA management. The approach presented here can be used as an early warning system to identify PAs where homogenization is occurring and alter management actions to halt this.
We must, however, also consider possible caveats associated with our approach. First, the Mann-Kendall-trend test must be interpreted as a measure of consistency of the trend and not the intensity of the trend. A maximum Mann-Kendall value can be reached with a small difference in absolute values (e.g., heterogeneity), if the change was constant over time. If the focus is on measuring the intensity of the trend, other analyses such as Structural Equation Models (SEMs) may be more informative. Secondly, ecological components and processes have an underlying spatial structure that is locally heterogeneous, for example topography, soil type, and watershed areas contribute to the overall complexity of the landscape (MacFadyen et al. 2016). For example, the two PAs with the highest mean Roa Q values were Songimvelo and Itala Game Reserve, which have high levels of topographical variability. Interpretations of heterogeneity should be cognisant of this – however, because we were interested in changes in heterogeneity over time, we assumed that these stable physical elements did not change over a 28-year period, and that the changes that were detected were linked to changes in vegetation as reflected by the spectral variation in the remote sensing product. Third, we did not include heterogeneity values for the years 2008 – 2012, because of the scan line correction error of Landsat 7. Although these regularities are subtle, they are still visible in the product and likely would have affected our calculations of heterogeneity. Nonetheless, we are confident that leaving out these values would not affect the overall trends reported here, given the length of the time series and the application of the Mann-Kendall-trend tests.